Wyniki wyszukiwania - Biblioteka Nauki

1

On combining principal components with Fisher's linear discriminants for supervised learning

100%

Pechenizkiy M. , Tsymbal A. , Puuronen S.

Foundations of Computing and Decision Sciences

|

2006

|

tom Vol. 31, No. 1

59-73

EN

"The curse of dimensionality" is pertinent to many learning algorithms, and it denotes the drastic increase of computational complexity and classification error in high dimensions. In this paper, principal component analysis (PCA). parametric feature extraction (FE) based on Fisher's linear discriminant analysis (LDA), and their combination as means of dimensionality reduction are analysed with respect to the performance of different classifiers. Three commonly used classifiers are taken for analysis: ŁNN, Naive Bayes and C4.5 decision tree. Recently, it has been argued that it is extremely important to use class information in FE for supervised learning (SL). However, LDA-based FE, although using class information, has a serious shortcoming due to its parametric nature. Namely, the number of extracted components cannot be more that the number of classes minus one. Besides, as it can be concluded from its name, LDA works mostly for linearly separable classes only. In this paper we study if it is possible to overcome these shortcomings adding the most significant principal components to the set of features extracted with LDA. In experiments on 21 benchmark datasets from UCI repository these two approaches (PCA and LDA) are compared with each other, and with their combination, for each classifier. Our results demonstrate that such a combination approach has certain potential, especially when applied for C4.5 decision tree learning. However, from the practical point of view the combination approach cannot be recommended for Naive Bayes since its behavior is very unstable on different datasets.

2

PCA-based approximation of a class of distributed parameter systems: classical vs. neural network approach

88%

Bartecki K.

|

tom Vol. 60, nr 3

651-660

EN

In this article, an approximation of the spatiotemporal response of a distributed parameter system (DPS) with the use of the principal component analysis (PCA) is considered. Based on a data obtained by the numerical solution of a set of partial differential equations, a PCA-based approximation procedure is performed. It consists in the projection of the original data into the subspace spanned by the eigenvectors of the data covariance matrix, corresponding to its highest eigenvalues. The presented approach is carried out using both the classical PCA method as well as two different neural network structures: two-layer feed-forward network with supervised learning (FF-PCA) and single-layer network with unsupervised, generalized Hebbian learning rule (GHA-PCA). In each case considered, the effect of the approximation model structure represented by the number of eigenvectors (or, in the neural case, units in the network projection layer) on the mean square approximation error of the spatiotemporal response and on the data compression ratio is analysed. As shown in the paper, the best approximation quality is obtained for the classical PCA method as well as for the FF-PCA neural approach. On the other hand, an adaptive learning method for the GHA-PCA network allows to use it in e.g. an on-line identification scheme.

3

Sparse Bayesian learning in classifying face feature vectors

88%

Momot A. , Kawulok M.

|

2005

|

tom Vol. 9

151--158

EN

The Relevance Vector Machine (RVM), a Bayesian treatment of generalized linear model of identical functional form to the Support Vector Machine (SVM), is the recently developed machine learning framework capable of building simple models from large sets of candidate features. The paper describes the application of the RVM to a classification algorithm of face feature vectors, obtained by Eigenfaces method. Moreover, the results of the RVM classification are compared with those obtained by using both the Support Vector Machine method and the method based on the Euclidean distance.

4

A strategy learning model for autonomous agents based on classification

88%

Śnieżyński B.

|

tom 25

|

nr 3

471-482

EN

In this paper we propose a strategy learning model for autonomous agents based on classification. In the literature, the most commonly used learning method in agent-based systems is reinforcement learning. In our opinion, classification can be considered a good alternative. This type of supervised learning can be used to generate a classifier that allows the agent to choose an appropriate action for execution. Experimental results show that this model can be successfully applied for strategy generation even if rewards are delayed. We compare the efficiency of the proposed model and reinforcement learning using the farmer-pest domain and configurations of various complexity. In complex environments, supervised learning can improve the performance of agents much faster that reinforcement learning. If an appropriate knowledge representation is used, the learned knowledge may be analyzed by humans, which allows tracking the learning process.

5

Stationary supercapacitor energy storage operation algorithm based on neural network learning system

88%

Jefimowski W. , Nikitenko A. , Drążek Z. , Wieczorek M.

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2020

|

tom Vol. 68, nr 4

733--738

EN

The paper proposes to apply an algorithm for predicting the minimum level of the state of charge (SoC) of stationary supercapacitor energy storage system operating in a DC traction substation, and for changing it over time. This is done to insure maximum energy recovery for trains while braking. The model of a supercapacitor energy storage system, its algorithms of operation and prediction of the minimum state of charge are described in detail; the main formulae, graphs and results of simulation are also provided. It is proposed to divide the SoC curve into equal periods of time during which the minimum states of charge remain constant. To predict the SoC level for the subsequent period, the learning algorithm based on the neural network could be used. Then, the minimum SoC could be based on two basic types of data: the first one is the time profile of the energy storage load during the previous period with the constant minimum SoC retained, while the second one relies on the trains’ locations and speed values in the previous period. It is proved that the use of variable minimum SoC ensures an increase of the energy volume recovered by approximately 10%. Optimum architecture and activation function of the neural network are also found.

6

Random generalization by feedforward neural networks

88%

Rataj A.

|

tom Vol. 20, No. 2

117-124

EN

A generalization by a feedforward neural network is discussed, such that some samples may be generalized using very dierent, conflicting criteria. A training set is deliberately constructed to show that feedforward neural networks in such a case can generalize very spuriously and randomly. To illustrate the dierences between dierent learning machines, results given by a small subset of the support vector machines are also presented.

PL

W artykule dyskutowane jest uczenie jednokierunkowych sieci neuronowych w którym niektóre próbki mogą być uogólnianie przy użyciu bardzo odmiennych kryteriów. Zbiór uczący jest specjalnie skonstruowany w sposób pokazujący, że w przypadku istnienia bardzo odmiennych kryteriów uogólniania sieć neuronowa może generalizować w sposób przypadkowy, wynikły prawie całkowicie ze struktury wewnętrznej sieci, a nie z zawartości pliku uczącego. Sieci neuronowe są porównane też do SVM-ów, które przy odpowiednich parametrach nie wykazały takiej losowości, jednak z drugiej strony w testowanych przypadkach nie potrafiły uogólnić niektórych wzorców w pliku uczącym.

7

Deep Neural Network for Supervised Single-Channel Speech Enhancement

75%

Saleem N. , Irfan Khattak M. , Ali M. Y. , Shafi M. , Saleem N. , Irfan Khattak M. , Ali M. Y. , Shafi M.

|

8

Improving the efficacy of automated fetal state assessment with fuzzy analysis of delivery outcome

63%

Czabanski R. , Jezewski M. , Horoba K. , Jezewski J. , Leski J.

|

tom Vol. 24

223--230

EN

A number of methods of the qualitative assessment of fetal heart rate (FHR) signals are based on supervised learning. The classification methods based on the supervised learning require a set of training recordings accompanied by the reference interpretation. In the real data collections the class of signals related to fetal distress is usually under-represented. Too small percentage of distress patterns adversely affects the effectiveness of the automated evaluation of the fetal state. The paper presents a method of equalizing the class sizes based on the reference assessment of the fetal state with the fuzzy analysis of the newborn attributes. The supervised learning with increased number of the FHR signals, which are characterized by the highest rate of the fuzzy inference leads to significant increase of the efficacy of the qualitative assessment of the fetal state using the Lagrangian support vector machine.

9

Comparison of supervised learning methods for spike time coding in spiking neural networks

63%

Kasiński A. , Ponulak F.

|

2006

|

tom Vol. 16, no 1

101-113

EN

In this review we focus our attention on supervised learning methods for spike time coding in Spiking Neural Networks (SNNs). This study is motivated by recent experimental results regarding information coding in biological neural systems, which suggest that precise timing of individual spikes may be essential for efficient computation in the brain. We are concerned with the fundamental question: What paradigms of neural temporal coding can be implemented with the recent learning methods? In order to answer this question, we discuss various approaches to the learning task considered. We shortly describe the particular learning algorithms and report the results of experiments. Finally, we discuss the properties, assumptions and limitations of each method. We complete this review with a comprehensive list of pointers to the literature.

10

Basic quantum circuits for classification and approximation tasks

63%

Wiśniewska J. , Sawerwain M. , Obuchowicz A.

|

tom Vol. 30, no. 4

733--744

EN

We discuss a quantum circuit construction designed for classification. The circuit is built of regularly placed elementary quantum gates, which implies the simplicity of the presented solution. The realization of the classification task is possible after the procedure of supervised learning which constitutes parameter optimization of Pauli gates. The process of learning can be performed by a physical quantum machine but also by simulation of quantum computation on a classical computer. The parameters of Pauli gates are selected by calculating changes in the gradient for different sets of these parameters. The proposed solution was successfully tested in binary classification and estimation of basic non-linear function values, e.g., the sine, the cosine, and the tangent. In both the cases, the circuit construction uses one or more identical unitary operations, and contains only two qubits and three quantum gates. This simplicity is a great advantage because it enables the practical implementation on quantum machines easily accessible in the nearest future.

11

Application of selected supervised classification methods to bank marketing campaign

63%

Grzonka D. , Borowik B. , Suchacka G.

|

tom Vol. 5, No. 1

36--48

EN

Supervised classification covers a number of data mining methods based on training data. These methods have been successfully applied to solve multi-criteria complex classification problems in many domains, including economical issues. In this paper we discuss features of some supervised classification methods based on decision trees and apply them to the direct marketing campaigns data of a Portuguese banking institution. We discuss and compare the following classification methods: decision trees, bagging, boosting, and random forests. A classification problem in our approach is defined in a scenario where a bank’s clients make decisions about the activation of their deposits. The obtained results are used for evaluating the effectiveness of the classification rules.

12

Application of artificial neural networks to define risk in construction process

51%

Dutkiewicz M.

|

tom nr 6

CD-CD

EN

The paper presents approach of application of neural network to support the decision in construction industry. Described method allowed to obtain a determination of the level of risk of not performance the investment on time using the parameters achievable at an early stage of the project. The model establishes a methodology that provides tools for calculating the risk. Conclusions for further investigations are attached.

PL

W artykule omówiono zastosowanie podejścia użycia sieci neuronowej stanowiącej wsparcie decyzji w budownictwie. Opisana metoda pozwoliła na określenie poziomu ryzyka niewykonania inwestycji na czas używając parametrów osiągalnych na etapie wczesnej realizacji projektu. Określono metodologię, która dostarcza środków do obliczania ryzyka. Przedstawiono wnioski dotyczące dalszych badań.

13

Human identification based on a kinematical data of a gait

51%

Świtoński A. , Mucha R. , Danowski D. , Mucha M. , Cieślar G. , Wojciechowski K. , Sieroń A.

|

tom R. 87, nr 12b

169--172

EN

The paper is devoted to the gait identification challenges. It evaluates human abilities to recognize gait on the basis of skeleton animations. Further, it proposes the method of gait identification based on the kinematical data .The feature extraction approach and supervised learning are applied. To explore the most individual joints movements, aggregated feature rankings are calculated. To examine the proposed method, the database containing 353 gaits of 25 different actors is collected in the motion capture laboratory. We have obtained 99.7% of classification accuracy.

PL

W artykule zaprezentowano eksperyment oceniający zdolności człowieka do rozpoznawania chodu oraz zaproponowano metodę identyfikacji chodu na podstawie danych kinematycznych. Bazuje ona na podejściu z ekstrakcją cech i uczeniem nadzorowanym. W celu oceny ruchu poszczególnych stawów pod kątem osobniczych cech różnicujących wyznaczono zagregowane rankingi atrybutów. Do weryfikacji zaproponowanej metody, zgromadzono bazę 353 przejść wykonywanych przez 25 różnych aktorów. Uzyskano ponad 99% skuteczność klasyfikacji.

14

Sztuczne sieci neuronowe i kierunki ich zastosowań w przemyśle. Inspiracje biologiczne sztucznych sieci neuronowych, ich historia oraz matematyczny model neuronu do zastosowań inżynierskich

51%

Drobina R. , Grecka A. , Lewandowski S.

|

tom nr 5

19--23

EN

Artificial intelligence is a branch of computer science who create computer programs that simulate intelligent human behavior. The main task of the study of artificial intelligence is designing machines and computer programs capable of carrying out certain functions of the mind and the human senses not amenable to simple numeric algorithmization. Particularly important are artificial neural networks useful to look for more complex relationships between input and output. A neural network is a mathematical paradigm modeling of biological activity and neutral system used to perform calculations. The article presents the biological inspirations and history of the development of artificial neural networks (ANN).

15

Sztuczne sieci neuronowe i kierunki ich zastosowań w przemyśle. Podstawowe rodzaje sztucznych sieci neuronowych

51%

Drobina R. , Grecka A. , Lewandowski S.

|

tom nr 7

26--31

EN

The article presents the basic types of artificial neural networks (ANN), designed to solve the regression problems, engineering applications, engineering manufacturing as well as in industrial conditions. The group included these networks are Adaline network, Madaline networks, linear, unidirectional network perpceptron type of multi-layer (MLP), Generalized Regression Neural Networks (GRNN) and a network of radial basis function (RBF).